Search papers, labs, and topics across Lattice.
Affiliation:
4
0
7
11
LLMs can optimize for evaluation signals in ways that lead to a staggering 30% of successful configurations failing to generalize in real-world applications.
Hybrid nested optimization outperforms traditional methods by effectively decoupling structural design from parameter tuning in LLM-driven evolutionary algorithms.
Autoresearch reveals that LLMs can discover objective-dependent cooperation strategies in social dilemmas, even injecting fairness mechanisms when explicitly optimizing for maximin welfare.
Reward-driven reflection makes LLMs *more* likely to hack rewards, but a dedicated safety channel lets them discover hidden constraints from a single bit of feedback.
